meraGPT - Small models for one job; Decider 1 leads typed-decisions

Hey Product Hunt 👋 meraGPT is an API for small models that each do one job. Restyler 1 takes text from any model and rewrites it so it reads the way a person writes. We measured it with a classifier trained to tell model output from human writing on the same prompts, where 0.5 means it can't tell and 1.0 means it's obvious. Claude Haiku 4.5, Gemini 3.5 Flash Lite and GPT-5.6 Luna all score about 0.90 on their own. After Restyler 1 the same text scores 0.65 to 0.70. Gemini was held out of training entirely and still improved the most. It keeps 73% of facts on average, and every fact in only 41% of blocks, so read the output wherever accuracy matters. It was not built or measured for getting past AI detectors. Query Fanout 1 takes a question and returns the searches an assistant would likely run before answering. It learned from searches we observed ChatGPT, Claude and Gemini actually make. If you care whether your pages get cited in AI answers, those searches are what decide it. Both are OpenAI-compatible and cost $0.05 per million input tokens and $0.15 per million output. The playgrounds work without an account. We'd love to hear what you would point them at, and which narrow job you'd want a model for next.